Selective attention and immediate memory in dichotic listening

Retrospective Theses and Dissertations 1969 Selective attention and immediate memory in dichotic listening James Richard Craig Iowa State University Follow this and additional works at: http://lib.dr.iastate.edu/rtd Part of the Experimental Analysis of Behavior Commons, and the Psychiatry and Psychology Commons Recommended Citation Craig, James Richard, "Selective attention and immediate memory in dichotic listening " (1969). Retrospective Theses and Dissertations. 3637. http://lib.dr.iastate.edu/rtd/3637 This Dissertation is brought to you for free and open access by Iowa State University Digital Repository. It has been accepted for inclusion in Retrospective Theses and Dissertations by an authorized administrator of Iowa State University Digital Repository. For more information, please contact digirep@iastate.edu.

This dissertation has been microiihned exactly as received 70-7685 CRAIG, James Richard, 1944- SELECTIVE ATTENTION AND IMMEDIATE MEMORY IN DICROTIC LISTENING. Iowa State University, Ph.D., 1969 Psychology, experimental University Microfilms, Inc., Ann Arbor, Michigan

SELECTIVE ATTENTION AND IMMEDIATE MEMORY IN DICROTIC LISTENING by James Richard Craig A Dissertation Submitted to the Graduate Faculty in Partial Fulfillment of The Requirements for the Degree of DOCTOR OF PHILOSOPHY Major Subject: Psychology Approved : Signature was redacted for privacy. r Work Signature was redacted for privacy. Head of Major Department- Signature was redacted for privacy. De of feaduat College Iowa State University Ames, Iowa 1969

ii TABLE OF CONTENTS Page INTRODUCTION PROBLEM METHOD RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES 1 12 16 24 40 50 51

1 INTRODUCTION One of the core problems of psychology for late nineteenth and early twentieth century psychologists was that of attention (Bakan, 1966). Even though most psychologists of the time agreed on the importance of the concept there were considerable differences of opinion as to how attention would be treated. For example, the structuralist camp as exemplified by Titchener equated attention with clearness of sensation. The basic problem of accounting for the selection of certain stimuli in consciousness was accomplished by the use of the attention concept which in turn was reduced to sensory clearness (Marx and Hillix, 1963). Functionalists such as William James on the other hand believed attention to be an active process whose function was selection. That is, the active organism directs its attention to specific stimuli and, thereby, determines the nature of its experience (James, 1890). Only those things people notice was thought to influence their psychological functioning (James, 1890). The subsequent theoretical approaches which followed structuralism and functionalism tended to either ignore the concept of attention or to relegate it to a position of insignificant concern. The supporters of behaviorism, gestalt psychology, psychoanalysis, and S-R learning theory for a variety of reasons were not concerned with the problem of attention (Bakan, 1966). Recent developments, however.

2 suggest that attention is once again being considered a primary problem for investigation in psychology (Bakan, 1966). More and more the organism is being seen as an active seeker and selector of information; as an active perceiver rather than as a passive receiver. The organism has only a limited information handling capacity and, therefore, must select stimuli he is going to focus on from among the many available. These selection processes are currently being viewed as constituting attention (Bakan, 1966). However, the exact nature of the limitation and of the underlying processes of selection has been the subject of considerable argument (e.g., Treisman and Geffin, 1967; Deutsch and Deutsch, 1967; Treisman, 1967; Lindsay, 1967; Treisman, 1969). One approach to the investigation of the problem of selective attention has been through the use of the dichotic listening (DL) technique listening to two simultaneously arriving messages independently presented to the two ears. Broadbent (1957, 1958) was one of the first to incorporate the existing DL and other data regarding attention and memory into a theoretical model. Broadbent's model of the human perceptual system proposes a system of limited capacity where information can be processed into memory in but one sensory channel (e.g., one ear at a time). Because of this limitation, a selective operation in the form of a "filter" mechanism is! performed upon all messages (input information) in 'the system.

3 The selective or "filter" operation takes the form of selecting a sensory channel based on the characteristics of the input information. (For example, the filter may be biased towards a given ear because of "inherent" perferences or habit factors.) The information in the selected channel is passed into what Broadbent calls a "perceptual" or "p" system. Tha incoming information in the other channel is blocked before reaching the central processing or "p" system. It is held at a stage prior to processing in a temporary, shortterm store or "s" system. Information stored in the "s" system is said to "leave the system" (to undergo autonomous decay) if it is held in that store too long before it is processed. In short, information arriving in one channel is immediately processed by the "p" system. The information in the other channel is blocked peripherally and shunted into the "s" system to be processed when the "p" system is free, providing the information is still available. The proposed filter operation has since been revised through the work of Treisman (1960, 1964, 1969) and supported by the findings of Broadbent and Gregory (1963). The Treisman revision suggests that the mechanism for selective attention is an attenuation process rather than a peripheral blocking of the irrelevant messages and that a hierarchy of tests is carried out on inputs entering all channels whether they are selected for conscious attention or not. That is, the

4 same pattern-recognition processes work on both relevant and irrelevant (as defined by E) information although the "raw data" on which the processes operate are degraded in the latter case. The decision criteria are the same for both channels, but the "strength" of the incoming information signals is lower in the secondary channel. The attentuated input signals in the secondary channel fail to "pass" tests relatively low in the decision hierarchy because of the "degraded" nature of the signals and, thus, the information goes unrecognized. The data supporting Broadbent's model (1957, 1958) and Treisman's revision (1960, 1964, 1967) have come largely from DL studies of shadowing repeating information aloud as it arrives in one channel while "ignoring" information arriving in the other channel or different information arriving in the same channel. Cherry (1953) found that Ss have difficulty in shadowing one of two dichotically presented passages but could relatively easily "reject" the unwanted passage. When S^s were asked afterwards what they had heard in the "rejected" or ignored channel, they could say little about it except that sounds of some type had been presented. Moray (1959) attempted to investigate more closely this finding by testing an S's ability to recall information presented to the ignored channel. Ss were asked to shadow a selection of prose while a list of words was presented to the other ear.

5 Thirty seconds after termination of the presentation, the S^s were unable to recognize the list that had been presented to the ignored channel. Treisman (1964) in an investigation of shadowing efficiency presented to one ear a one minute selection of prose which was to be shadowed (the primary channel). She presented 0> 1, or 2 different irrelevant selections either to the other ear alone (the secondary channel) or to both ears simultaneously. She found that shadowing performance was largely determined by the number of channels to be rejected (whether the two irrelevant messages were separate or mixed on one channel) rather than simply the difficulty or ease in separating the selections. She also found that different types of stimuli presented to the secondary channel (prose, numbers, Czechoslovakian, or random English words) had no significant effect on shadowing performance. Treisman concluded that this confirmed her earlier findings (1960) that Ss remain almost completely unaware of the content of the secondary channel in dichotic listening. Thus, in a review of the general findings in the area, Egeth (1967) reached the conclusion that the ignored message has little, if any, behavioral effect. It must be noted, however, that the S^s in the studies cited above were not told prior to the presentation of the information that the input to the secondary channel would be

6 of any importance or that they would later be asked questions about the input. Also, the test for memory was usually delayed for a definite period of time of up to 30 seconds after presentation. Such a delay would not be an adequate test for short-term memory of the secondary channel information. Considerable support for the attenuation hypothesis of selective attention in DL was found by Broadbent and Gregory (1963) based on a signal detection analysis. Briefly, the basic idea behind this type of analysis is that an individual's performance is a function not only of his sensitivity but also dependent on his relative willingness to make particular responses (Green and Swets, 1965). The distributions of responses when a signal is present or absent are theorized to take the form of overlapping normal distributions of equal variance. It is further assumed that is able to convert a priori probabilities of occurrence of signal and of noise as well as the costs and values of specific responses to a monotonie scale. Given these assumptions it is possible to compute two independent measures. One measure, called sensitivity or d', is the distance between the means of the signal and noise distributions, m^-m^, divided by the common standard deviation (Green and Swets, 1966). The second measure, called response bias or Beta, indicates where on the monotonie scale the criterion point has been placed

7 relative to the point where the signal and noise distributions intersect (Green and Swets, 1966). Thus, the response bias measure yeilds an index as to where the stops saying "no" and begins responding "yes". Broadbent and Gregory (1963) presented to one ear (the secondary channel) of Ss short bursts of noise which might or might not have contained a tone. The other ear (the primary channel) received six digits. The Ss reported their degree of confidence that a tone was presented to the secondary channel under the two conditions of either ignoring the numbers or of "shadowing" the numbers while making the judgements about the tone. The results showed that the signal detection response bias measure remained relatively unchanged by division of attention. They did find that the sensitivity measure d' changed when attention was divided. From these findings they concluded that division of attention away from a stimulus produces an effect resembling a reduction in the intensity of a stimulus. The ignored event is not blocked altogether and under suitable conditions may produce a response from an observer a conclusion generally taken as supporting the Treisman {I960, 1964) interpretation of attenuation of input rather than peripheral blocking. An alternative view of the selective attention process has been expressed by Deutsch and Deutsch (1963). They hypothesize that incoming messages reach the same perceptual

8 and discriminatory mechanisms whether conscious attention is paid them or not. Such information is then grouped or segregated by these central mechanisms. How such a grouping takes place they do not suggest. However, they do hypothesize that each central structure which is excited by the presentation of a specific quality or attribute to the senses is given a pre-set weighting of importance which varies over time. The central structure or classifying mechanism with the highest weighting will transfer this weighting to the other classifying mechanisms with which it has been grouped. Thus, the information handled by those structures given the greatest weighting will be consciously attended to. A somewhat more detailed central processing view of selective attention is offered by Norman (1968). His ideas are advanced on the basic premise that selective attention is a partially automatic process. That is, selection among various sources or types of information occurs only after a certain amount of automatic information processing has already taken place. The assumption is made that information activates its image in memory storage without intervening cognitive processes being necessary. Memory is important in the selection process in that it provides information which serves as a model against which incoming sensory information is compared and analyzed. Given these basic assumptions, selection is hypothesized to proceed in the

9 following manner. Incoming sensory information is first processed in terms of its physical characteristics. The special features of the signal as determined by this processing determines the appropriate addresses in memory of the information relevant to the current input. The relevant representations in memory are then excited by the incoming information. Simultaneously, higher-level cognitive processes determine which stored representations are most important to current on-going psychological processes. These cognitive processes are collectively labeled pertinence by Norman (1968). Primarily, what is meant by pertinence is the expectations of the organism of future inputs and the particular properties of current information. Pertinence determination proceeds or occurs at the same time information is being analyzed and not after selection has taken place. It activates or labels certain information as being of importance to the organism on the basis of current psychological functioning. What information is selected for further processing by the attention mechanism is determined by which stored representations receive the greatest combination of pertinence and sensory activation. That is, selection of outputs of the storage system is determined by which information receives the highest levels of combined activation. Incoming information which is not selected "fades" or "decays"

10 quite quickly so that only a very sketchy or bare outline is left after a few seconds. Thus, the difference between the theories offered by Deutsch and Deutsch (196 3) and Norman (1968) on the one hand and by Broadbent (1957, 1958) with the Treisman revision (1960, 1964, 1969) on the other is that the selective attention process is hypothesized to be a central decision process rather than a peripheral attenuation process. In signal detection theory terminology Deutsch and Deutsch and Norman would appear to be saying that the selective attention process is not solely one of decreasing sensitivity to sensory input, but rather primarily one of decision processes determining which information is of importance and is to be selected for further processing. Although several previously cited studies (Cherry, 1953; Moray, 1959; Treisman, 1960, 1964; Egeth, 1967) indicated information presented to the secondary channel in DL was not remembered, other studies have found contradictory results. Mowbray (1964) and Peterson and Kroener (1964) found that when Ss were instructed beforehand to recall information presented to the secondary channel, they could, in fact, do so with a relatively high degree of accuracy. Norman (1969) has also demonstrated Ss do remember material presented to the secondary channel. He required Ss to shadow words presented to one ear at the rate of two words per second

11 and tested the S^s ' memory for two-digit numbers presented to the other ear after various periods of time between presentation and recall. He found when s were tested immediately after presentation, they did remember some of the digits. When S^s had to continue shadowing for 20 seconds before trying to remember the digits (as was done by Moray (1959) and others), the Ss could remember none of the digits which had been presented. Furthermore, Lewis (1968) found evidence against the attenuation hypothesis in DL when reaction times to words being shadowed differed as a function of the type of words presented to the secondary channel. He found that the meaningfulness of the different types of words in the secondary channel directly affected reaction times to the words to be shadowed. Thus, he concluded that even when the content of the message presented to the secondary channel cannot be reported by s, its meaning does affect S^s ' behavior significantly. This finding would suggest that information presented to the secondary channel in DL is not "filtered" or attenuated at a peripheral level.

12 PROBLEM The major purpose of this research was to attempt to set up a DL situation where the predictions of the theoretical position of Norman (1969) and Deutsch and Deutsch (1963) were different from those offered by Treisman's revision (1964, 1967) of Broadbent (1957, 1958). This was accomplished by employing a DL selective attention situation very similar to that used by Broadbent and Gregory (1963) but which employed the methods and techniques more commonly used in DL and recognition memory studies. Specifically, the Broadbent and Gregory (1963) study contained several methodological differences from other work in the area which could possibly have affected the results obtained. First, the type of "shadowing" asked of the S^s was different than that usually employed. Broadbent and Gregory had their s "shadow" digits by listening to a string of six and then writing them down. The task appears to have been more of a short-term memory type task than the usual verbal shadowing task (repeating items aloud as they are heard). Furthermore, Broadbent and Gregory do not report the accuracy with which their ^s "shadowed" the digits. Thus, they offer no behavioral evidence of how closely the Ss were attending to and shadowing the primary channel. Another methodological difference was the type of memory or detection task asked of the S^s for information presented to the secondary channel.

13 Presenting tones against a background of white noise lends itself to the signal detection analysis they used, but it does not coincide with the experimental situation employed in the majority of the DL studies. A finding growing out of the methodological orientation of most DL studies has been what is called ear asymmetry. Specifically, researchers have found when Ss are dichotically presented pairs of items (either digits or words), the information presented to the right ear is remembered better than that presented to the left for right-handed Ss (Satz, Achenbach, Pattishall, and Fennell, 1965; Bryden, 1964; Kimura, 1961; Borkowski, Spreen, and Stutz, 1965; Bartz, Satz, Fennell, and Lally, 1967). Although ear asymmetry is usually interpreted in terms of certain physiological attributes (Roberts, 1964; Satz, Achenbach, Pattishall, and Fennell, 1965), recent evidence (Johnson, 1967) suggests ear asymmetry may be the result of a habit factor since it appears to disappear with practice. Nonetheless, ear asymmetry is a factor that should be controlled for in DL experimentation. Since Broadbent and Gregory (1963) failed to do so, the present study controlled for possible differences in ears by giving Ss a considerable amount of practice in the task and by counterbalancing among the S_s which ear was shadowed first in the experimental manipulations.

14 Therefore, this experiment employed a DL situation which was more in line with techniques and procedures used in other DL studies. Instead of Ss listening to a string of items, remembering them, and then writing them down, the more usual shadowing technique of having s repeat items aloud as they are presented was used. Also, the stimulus materials were all from the spoken English language. The method used to test for detection and memory for items presented to the secondary ear was a recognition procedure. After the presentation of a series of items, a test item was given and S was to decide whether or not that particular item had just been presented. The use of this method of testing for memory allowed the use of the signal detection theory measures which separates each S^'s responses into a sensitivity component and a response bias component. Another question of interest was the effect that verbal shadowing has on the performance observed in DL shadowing studies. Perhaps verbal shadowing (repeating items aloud as they are heard) may constitute two listening tasks for the instead of one. The must not only monitor and repeat aloud information presented to one ear, but he must also generate and listen to his own reproductions of that information. Possibly, using another type of shadowing task may yield somewhat different findings. Therefore, a "motor" shadowing task was devised wherein the S was asked

15 to shadow the information in one channel by moving the appropriately labeled toggle switch on a display panel in front of him. Although the motor shadowing task also required monitoring of reproduction performance by the S^s to assure the desired response has been made, the monitoring task does not involve a "second" listening task. If the Broadbent and Gregory (1963) data were to be replicated, it was expected that the signal detection sensitivity measure would show more sensitivity when s weire asked to both shadow and do the recognition task. The signal detection response bias measure was not expected to change if the Broadbent and Gregory data were applicable. If the Norman (1969) and Deutsch and Deutsch (1963) theoretical position was to be supported, the opposite findings were expected. Sensitivity was not expected to change whether shadowing or not. But, response bias was expected to change with shadowing conditions. No prediction of the results of the comparison between motor and verbal shadowing was made.

16 METHOD Subjects Six Iowa State University undergraduate volunteers were paid for serving as S^s. Each S^ was right-handed and reported no history of hearing defects. Handedness was determined both by verbal report of each 2 by E observing which hand was used to write by each S. Stimulus materials The Ss were dichotically presented six pairs of items. The first five items presented to one ear were all letters and were chosen randomly without replacement from the following set of letters B, C, F, Gr H, K, M, Q, R, and S. The letters were chosen from the Conrad (1964) norms to attempt to keep acoustic confusions at a minimum. The first five items presented to the other ear were all color names randomly chosen without replacement from a set of ten BEIGE, BLUE, BROWN, GREEN, GOLD, GREY, ORANGE, PINK, RED, and TAN. The last items presented to each ear were recognition test items from the opposite class of stimuli. The trials were dichotically recorded by first marking a recording tape with a clearly visible mark at what would be one second intervals. The recorder was operated at the speed of 3 and 3/4 inches per second so a group of six marks were made at intervals of 3 and 3/4 inches on the recording

17 tape. A gap of 22 and 1/2 inches (six seconds) was left between each group of six marks. Then, one channel of the tape was recorded by speaking an item when one of the marks on the recording tape passed a fixed point on the recorder. The other channel was also recorded using the same procedure. The items were checked for simultaneity of presentation by playing the recording and observing the VU meters for both channels of the recorder. If the indicating needles of the VU meters appeared to E to move simultaneously with -. the onset of items in each channel, the trial was used as originally recorded. If the indicating needles of the VU meters appeared to move at different times, the trial was rerecorded until simultaneity was observed. Four tape recordings were made using this procedure. After one recording was recorded and checked, a duplicate recording was recorded off of the marked tape. The marked tape was then erased and used to make another recording. Design Six pairs of items were presented dichotically at the rate of one pair per second. Immediately following the fifth pair of items, a pair of test items was presented. The last item presented to the ear to be shadowed was the test item for that particular experimental trial and of the opposite stimulus type. For example, if S were shadowing

18 letters the last item presented to that ear would be a color name. S would then respond indicating whether this (last) color name had appeared previously in the list presented to the opposite ear. The S^s were instructed to ignore the last item presented to the ear to be monitored. Constructing the tapes by recording two test items for each trial cut in half the number of tapes needed. The type of item to be shadowed was counter-balanced among the trials so that half of the time Ss were shadowing letters and monitoring color names while the other half of the time they were shadowing color names and monitoring letters. Three types of shadowing conditions were employed. Under one set of conditions (verbal shadowing), Ss were instructed to verbally shadow aloud items presented to one ear and to monitor the items presented to the other. In a second set of conditions the S^s were asked not to shadow at all, but only to monitor one ear and make the recognition decision (the no shadowing condition). For the third task, the motor shadowing task, S^s were seated in front of a display panel which contained a row of ten toggle switches. Above each toggle switch were two labels: a letter and a color name. Both were typed in capital pica type characters. The letters and color names were both arranged in alphabetical order from left to right. To indicate a particular item had been heard, the S switched a SPDT toggle switch to its opposite position.

19 Moving the switch turned on a small light visible only to E and indicated that that particular item had been shadowed. E then moved a second SPDT toggle switch on his side of the panel which switched off the light thereby making that particular circuit ready for the next trial. In addition to the shadowing task, the S^s were to monitor the items presented to the opposite ear. Recognition for items presented to the monitored ear was tested by presenting one item (a test item) and asking the S^s to decide whether or not that particular item had been heard. In order to obtain the signal detection measures of sensitivity and response bias Ss were asked to make a "yes-no" response as to whether or not a test item had been presented and to simultaneously give a certainty rating of their response. This was accomplished by having the S^s respond using a 1 to 99 rating scale where; 1 to 49 meant "no" the test item had not been presented with progressively less confidence, 50 was a "don't know" response, and 51 to 99 indicated "yes" the test item had been presented with progressively increasing confidence. A six second inter-trial interval was allowed for S^s to record their rating response with more time given if needed by stopping the tape recorder. The probability of occurrence of the test item in the ear to be monitored was.50 with each of the five serial positions being tested an equal number of times. The trials

20 consisted of 14 tests for each of the five serial positions of the items to be monitored when the test item was present; seven trials were composed of color names to be monitored and seven made up of letters. Seventy trials were recorded when the test item was not presented to the ear to be monitored. Thus, 140 experimental trials were used and the presence or absence of the test items in the monitored ear was randomly ordered among the 140 trials. Four such sets of 140 trials were recorded. One set was used exclusively for practicing the Ss in the various tasks. The three other sets of 140 trials were used for the experimental trials one for each of the three different shadowing conditions. Each of the three tapes were randomly assigned to two Ss in each of the shadowing conditions so that the particular arrangement of stimuli on the separate tapes was not confounded with shadowing conditions. Each S shadowed with one ear once through the set of 140 trials and then shadowed the other ear through the same 140 trials after a three to five minute break. Whether the right or left ear was shadowed first was counterbalanced among the six s as well as among the three sets of experimental trials. Thus, the design was a within factorial design across six Ss: 3 (type of shadowing: verbal, motor, or no shadowing) X 2 (ears: shadowing the right or the left ear) X 2 (stimulus materials: shadowing letters or color names) X

21 5 (serial position: serial positions 1 through 5) X 2 (present or absent: whether the test item was or was not presented to the ear being monitored) X 7 (number of observations per cell). Procedure The tape recordings were recorded and played on a Wollensak 5730 stereo tape recorder and were presented to each via Koss Pro-4 headphones. The basic task of the Ss was to shadow the items presented to one ear and to monitor the items presented to the other. The last item presented to the shadowed ear was the test item for the monitored ear on each trial. Each S was to decide whether or not that particular item had just been presented to the monitored ear and to simultaneously give a certainty rating of his response. Each received 2 onehour practice sessions of verbal shadowing and 1 one-hour practice session in the motor shadowing task. One set of 140 trials was used exclusively for all practice sessions. The S_s were instructed to shadow as well as they possibly could and were told their pay would be determined by their shadowing performance. That is, they were told their wage would be based on a 95% correct shadowing rate and, that if their shadowing performance dropped below 95%, their wage would decrease proportionately. However, only data from

22 trials where S^s made no mistakes in shadowing were, in fact, included in the analysis. Therefore, when errors in shadowing were made, extra trials randomly selected from another tape were given until a full set of the appropriate 140 trials was obtained. (Most Ss were shadowing so accurately that only a few extra trials were needed.) One experimental session consisted of 280 trials in two blocks of 140 trials each. Each block of 140 was separated by a rest interval of three to five minutes. The S shadowed the first block of trials with one ear and the second block with the other. Each S participated in three practice sessions (two for verbal shadowing and one for motor shadowing) and three experimental sessions (one for each of the shadowing conditions). Scoring and analysis Shadowing performance was scored by E by marking trials where errors were made. Omissions, mispronounciations, and shadowing words out of order were considered to constitute errors in shadowing. The S^s' rating responses were transformed to control for the tendency of individuals to make finer discriminations between successive points towards the middle of the scale than towards the extremes. The transformation used was the arc sine t/p transformation which has the effect of homogenizing

23 the scale. The proportions necessary to use the transformation were obtained by simply dividing each rating obtained from the 99 point scale by 100. Average percentages of correct recognitions were also computed for several of the manipulations. This was defined as the combined average percentage of saying "yes" when the test item had been presented and of saying "no" when the test item had not been presented. A 3 (type of shadowing) X 2 (ears) X 2 (type of stimulus materials) X 5 (serial position) analysis of variance was conducted separately for when the test item was present and when it was absent for each S. A signal detection type of analysis was also conducted so as to separate each S's responses into a sensitivity and a response bias component. This will be described in detail later.

24 RESULTS The results of the analyses of variance are summarized in Table 1. Examination of Table 2 which summarizes the Table 1. Summary of analyses of variance for each S^ ANALYSIS OF VARIANCE FOR S^ - TEST ITEM PRESENT Source df MS F P A (type of shadowing) 2 2.062 24.498 <.01 B (ears) 1 0.001.006 C (stimulus materials) 1 2.324 27.614 <.01 D (serial position) 4.101 1.198 AB 2.311 3.698 AC 2.443 5.262 <.01 AD 8.226 2.681 BC 1.016.193 BD 4.109 1.289 CD 4.073.864 ABC 2.772 9.167 ABD 8.148 1.757 <.01 ACD 8.076.898 BCD 4.014.170 Error 368.084 TOTAL 419 ANALYSIS OF VARIANCE FOR S^ - TEST ITEM ABSENT Source df MS F P A 2 9.467 214.362 <.01 B 1.012.261 C 1.026.594 D 4.035.784 AB 2.053 1.200 AC 2.074 1.681 AD 8.028.628 BC 1.013.286 BD 4.029.664 CD 4.024.543 ABC 2.032.716 ABD 8.036.815

25 Table 1 (Continued) Source df MS ACD BCD Error TOTAL 8 4 368 419.022.035.044.499.784 ANALYSIS OF VARIANCE FOR S^ - TEST ITEM PRESENT Source df MS F A (type of shadowing) 2 2.731 25.660 B (ears) 1.132 1.238 C (stimulus materials) 1.061.576 D (serial position) 4 2.164 20.333 AB 2.466 4.382 AC 2.066.619 AD 8.425 3.994 BC 1.016.148 BD 4.084.786 CD 4.073.690 ABC 2.059.557 ABD 8.104.980 ACD 8.205 1.925 BCD 4.218 2.048 Error 368.106 TOTAL 419 ANALYSIS OF VARIANCE FOR S^ - TEST ITEM ABSENT Source df MS F P A 2 10.746 133.451 B 1.032.400 C 1 2.014 25.008 D 4.026.325 AB 2.478 5.938 <.01 AC 2.160 1.991 AD 8.080.996 BC 1.0002.003 BD 4.022.279 CD 4.036.442 ABC 2.009.113 ABD 8.042.520 ACD 8.037.465 BCD 4.078.974 Error 368.081 TOTAL 419 V V 1 1 o 1 1 o

26 Table 1 (Continued) ANALYSIS OF VARIANCE FOR S^ - TEST ITEM PRESENT Source df MS F P A (type of shadowing) 2 3.654 24.803 <.01 B (ears) 1.020.134 C (stimulus materials) 1 2.227 15.118 <.01 D (serial position) 4 1.152 7.820 <.01 AB 2.436 2.960 AC 2.119.808 AD 8.225 1.525 BC 1.083.566 BD 4.163 1.110 CD 4.410 2.781 ABC 2.318 2.159 ABD 8.084.569 ACD 8.-18.800 BCD 4.334 2.266 Error 3G8.147 TOTAL 419 ANALYSIS OF VARIANCE FOR S^ - TEST ITEM ABSENT Source df MS F A 2 6.234 36.882 B 1.232 1.374 C 1.054.318 D 4.059.351 AB 2 1.155 6.832 AC 2.900 5.323 AD 8.032.192 BC 1.479 2.832 BD 4.017.101 CD 4.017.100 ABC 2.074.435 ABD 8.069.408 ACD 8.084.497 BCD 4.052.306 Error 368.169 TOTAL 419

27 Table 1 (Continued) Source df A (type of shadowing) 2 B (ears) 1 C (stimulus materials) 1 D (serial position) 4 AB 2 AC 2 AD 8 BC 1 BD 4 CD 4 ABC 2 ABD 8 ACD 8 BCD 4 Error 368 TOTAL 419 MS F 4.027 36.269 <.01.106.850.955 7.653 <.01 1.115 10.038 <.01.181 1.634.096.867.746 6.718.071.636.093.841.072.649.610 5.490 <.01.056.504.116 1.048.078.702.111 A O H ANALYSIS OF VARIANCE FOR S^ - TEST ITEM ABSENT Source df MS F P A 2 8.352 81.977 <.01 B 1.889 8.723 <.01 C 1.878 8.615 <.01 D 4.062.608 AB 2.033.327 AC 2.319 3.132 AD 8.058.569 BC 1.007.064 BD 4.028.270 CD 4.128 1.252 ABC 2.006.064 ABD 8.053.521 ACD 8.100.979 BCD 4.172 1.689 Error 368.102 TOTAL 419

28 Table 1 (Continued) ANALYSIS OF VARIANCE FOR S^ - TEST ITEM PRESENT Source df MS F P A (type of shadowing) 2 1.216 22.577 <.01 B (ears) 1.848 15.754 <.01 C (stimulus materials) 1.483 8.973 <.01 D (serial position) 4 1.786 33.181 <.01 AB 2.195 3.616 AC 2.001.023 AD 8.105 1.942 BC 1.029.548 BD 4.012.224 CD 4.019.355 ABC 2.105 1.955 ABD 8.095 1.769 ACD 8.073 1.359 BCD 4.073 1.351 Error 368 TOTAL 419 ANALYSIS OF VARIANCE FOR - TEST ' ITEM ABSENT ^5 Source df MS F P A 2 2.084 78.040 <.01 B 1.110 4.136 C 1.137 5.139 <.01 D 4.008.296 AB 2.101 3.778 AC 2.009.322 AD 8. 028 1.034 BC 1.219 8.196 BD 4.018.677 CD 4.080 2.978 ABC 2.014.523 ABD 8.007.275 ACD 8.025.926 BCD 4.042 1.586 Error 368.027 TOTAL 419 V i 1 o

29 Table 1 (Continued) ANALYSIS OF VARIANCE FOR S^ TEST ITEM PRESENT 6 Source df MS F P A (type of shadowing) 2 9.220 139.386 <.01 B (ears) 1.307 4.641 C (stimulus materials) 1 1.524 23.034 <.01 D (serial position) 4 1.798 27.174 <.01 AB 2.067 1.020 AC 2.166 2.512 AD 8.194 2.935 BC 1.006.087 BD 4.060.914 CD 4.008.120 ABC 2.016.692 ABD 8.223 3.375 ACD 8.042.638 BCD 4.067 1.020 Error 368.066 TOTAL 419 ANALYSIS OF VARIANCE FOR S, TEST ITEM ABSENT D Source df MS F P A 2 6.243 118.156 <.01 B 1.192 3.627 C 1,010.181 D 4.082 1.543 AB 2.173 3.275 AC 2.018.349 AD 8.062 1.164 BC 1.001.012 BD 4.018.342 CD 4.025.427 ABC 2.024.464 ABD 8.042.792 ACD 8.016.312 BCD 4.039.737 Error 368.053 TOTAL 419

30 Table 2. The error mean squares and standard deviations for each S for both present and absent conditions TEST ITEM PRESENT TEST ITEM ABSENT S EMS S EMS S 1.0842.290.0442.210 2.1064.326.0805.284 3.1473.384.1690.411 4.1110.333.1019.319 5.0538.232.0267.163 6.0662.257.0528.230 error mean squares of the analyses for each indicated it would be inadvisable to pool Ss for a combined analysis. These variance estimates showed that Ss varied considerably among themselves in their use of the rating scale. Furthermore, two s varied considerably in their use of the scale when the test item was present as compared to when it was absent while others showed smaller differences. The statistic described in Winer (1962) for combining several independent tests of the same hypothesis was used to obtain an over-all probability statement concerning the significance of the various main effects and interactions for test item present and absent. The inference associated with this statistic, may, at times, be different from the usual inference associated with significance tests. For example, the test for relative sensitivity of the two ears will be significant either if each responds similarly or if the more sensitive ear is conditional on the S^. Using the statistic described above for combining the

31 S^s' data (Winer, 1962), the type of shadowing task was found to interact with which particular ear was shadowed to significantly affect performance, x. (12) = 23.81, <.05, and 2 ^ (12) = 35.15, <.01, for the present and absent conditions respectively. Generally, small differences in recognition performance between the ears were noted under the motor and no shadowing conditions. This is reflected in the average percentage of correct recognitions which is the average percentage of saying "yes" when the test item was present and of saying "no" when the test item was absent. These percentages were 75% and 73% on the motor shadowing task and 92% and 93% under the no shadowing condition for the left and right ears respectively. On the verbal shadowing task recognition performance was higher and the s showed somewhat greater confidence in their responses when shadowing the left ear and monitoring the right than vice versa. The average percentage of correct recognitions showed that 76% of the items were correctly recognized when monitoring with the right ear as compared with 70% when monitoring with the left ear. The main effect of type of shadowing task was signifi- 2 cant across Ss, x. (12) = 41.55, <.01, for both present and absent conditions. The Ss generally displayed greater confidence in their responses and a higher average percentage of correct recognitions when they did not have to

32 shadow. There were no large, consistent differences in performance between the verbal and the motor shadowing tasks when summing across ears. Across the shadowing conditions, a small ear asymmetry effect was noted when the test item 2 was absent, ^ (12) = 24.22, p<.05. Ss tended to exhibit somewhat greater confidence and more accurate recognition performance when shadowing the left ear and monitoring the right ear than in the opposite case when the item was absent. The type of stimulus materials being shadowed (or monitored) was found to differentially affect performance 2 across Ss, ^ (12) = 66.31, <.01, for the present condition 2 and X. (12) = 33.33, <.01, for absent. The average correct recognition percentage was found to be 77% when color names were shadowed and letters were monitored. When letters were shadowed and color names monitored, the average correct recognition percentage was 83%. Also, the Ss tended to display more confidence in their responses by using a greater number of extreme ratings when shadowing letters (monitoring color names) than when shadowing color names (monitoring letters). The position of the test item (when it was presented) in the series of five items presented to the ear to be 2 monitored definitely affected performance, x. (12) = 69.91, <.01. (In the test item absent analyses ratings were

33 randomly assigned to the various "positions" for purposes of the signal detection analyses to be described later.) The performance of all Ss was found to show both a primacy and a recency effect. Specifically, there was a tendency to use higher ratings for serial position 1 than for serial position 2, a primacy effect. For each serial position succeeding position 2, the ratings were found to gradually increase. A tendency for some S^s to use lower ratings for the last serial position than for serial position 4 was found. The serial position effect was also reflected in the average percentages of correct recognitions (when the item was present only) for each serial position. The percentage of 69% for serial position 1 compared to 66% for position 2 reflects a small primacy effect. The percentages of 78%, 86%, and 84% for serial positions 3, 4, and 5 respectively show an increase in performance with serial position with a small decline in performance on position 5 also being noted. The second method of analysis proposed was to have been the usual signal detection analysis for recognition data. The assumptions of normal distributions of signal and noise and of equal variances of the distributions are basic to the application of the signal detection analysis incorporating likelihood ratios. However, the assumption of equal variances could not be met by the present data. For two of the s there

34 were marked differences in the variances of their ratings when the test item was absent (noise) as compared to when the test item was present (signal). This was determined by looking at the error mean squares of the analyses of variance for each and comparing these variance estimates for test item present and absent. These variances are shown in Table 2. Since the distributions of ratings for test item present and absent were not equally variable for all the S^s, an index of sensitivity and of response bias which took these inequalities into account was considered to yield the most appropriate information. The sensitivity measure d' is - m usually defined as: (Green and Swets, 1966) which m m is the same as : s where a = a. When o_ ^ a, n s n s n an index of relative sensitivity can be determined by: ra m - where s and s are estimates of the variance in =n s n responses for the signal and noise distributions respectively. By subtracting the two ratios one is effectively determining the area between the cumulative probability distributions of the ratings both when the test item was present and when n^s % it was absent. By adding +, a relative estimate of s n where along the various rating scale levels the changed his criteria or biases for responding. These indexes of sensitivity (called d* here) and of response bias (called

35 BetSg here) are relative indexes directly based on the variability of rating responses of the Ss. This does not allow direct arithmetic comparisons among s across conditions for the indexes, but relative trends or changes in the indexes across experimental manipulations are possible. And, since the indexes are directly related to variability of responding, the absolute size of the measures will change with the variability of the two distributions. Estimating the variance for either distribution across several factors may yield estimates which result in negative d* indexes. However, if enough observations per cell are (had been) available, variance estimates based on each cell would not (have) yield(ed) such negative indexes. Greater sensitivity is reflected in the d* index by larger numbers. Larger numbers with the Beta2 index reflect a greater willingness to respond "yes". Using-the d* and Betag measures derived above, the sensitivity and response bias indexes were applied to several of the experimental manipulations. As has been pointed out elsewhere (Raser, 1969) the sensitivity index yielded the same general findings as the analyses of variance previously described. The d* measure showed that for the shadowing task by ears interaction, the sensitivity of the two ears was relatively the same on the motor and no shadowing tasks as is shown in Table 3. Under the verbal shadowing condition

36 considerably more sensitivity was generally noted when Ss were monitoring with the right ear than with the left. Also, Table 3. The sensitivity measure across s for the type of shadowing task (motor, verbal, and no) by ears (left and right) interaction MOTOR.VERBAL NO S L R L R L R 1. 29.16 -.05.20 3.03 2.53 2.68.91 1.35.59 2.98 3.64 3 1.28.92 1.77 1.43 2.68 3.14 4 1.20 1.19.26 -.10 2.89 2.27 5 -.76 -.42 1.41.81 2.40 1.71 6.83.90.81.99 3.86 4.74 d*.49.56.92.65 2.97 3.01 sensitivity generally increased somewhat on the verbal task compared to the motor task. A large increase in sensitivity was found on the no shadowing task as compared to the other shadowing tasks. The d* index also indicated s were more sensitive to the monitored ear when shadowing letters and monitoring color names than vice versa. The sensitivity measures for each S over serial positions indicated the same general findings as shown by the analyses of variance as is shown in Table 4. The d* index exhibited a bow-shaped serial position curve for most S^s with some much

37 Table 4. The for sensitivity values (d*) each S over serial position 1 2 3 Ss 4 5 6 d* = 1.76 1.25 1.78.90.56 1.56 1.14 0 H 2.87 1.24 1.50.89.28 1.42 1.03 0 3 a «^ ^ 5 U).83 1.64 1.72 1.16.93 1.81 1.35.90 2.07 2.20 1.49 1.36 2.54 1.76 1.04 3.24 2.15 1.79 1.85 2.59 2.10 more bow shaped than others. Some ^ showed little difference in sensitivity for serial positions 4 and 5 while others showed noticeable differences. Again, the same descriptive account obtained via the analyses of variance. The d* index did reflect one finding that was not evident from the analyses of variance however. Several of the Ss exhibited almost equal sensitivity to the first two serial positions while clear differences in sensitivity were found for other Ss. The response bias or Betag measure did reflect different trends of bias for the ears across the three shadowing conditions as is shown in Table 5. On the motor and the no shadowing tasks ^s generally showed a somewhat greater willingness to say "yes" when shadowing the left ear and monitoring the right. In contrast, under the verbal shadowing condition Ss generally showed a greater willingness to say

38 Table 5. The response bias measure across Ss for the type of shadowing task (motor, verbal,and no) by ears (left and right) interaction MOTOR VERBAL NO S L R L R L R 1 6.55 7.01 6.73 6.65 5.42 5.22 2 5.59 5.49 5.28 5.59 4.40 4.69 3 4.04 4.86 4.45 4.19 4.30 4.19 4 6.65 5.59 4.19 5.03 7.16 6.07 5 5.42 4.40 4.30 4.51 7.61 6.28 6 5.22 4.69 4.19 4.77 7.91 6.21 Betag 5.58 5.34. 4.86 5.12 6.13 5.44 "yes" when shadowing the right ear and monitoring the left. Across the shadowing conditions mixed changes in response bias were found. Generally, Ss were more willing to say "yes" on the no shadowing task than under the other shadowing conditions. There was some tendency toward S^s saying "yes" more often on the motor task than on the verbal task but the trend was not wholly consistent. Betag changes associated with which type of stimulus material being shadowed were mixed. Almost all s showed either no bias or a greater willingness to say "yes" when shadowing letters. Response bias over serial positions generally exhibited a bow-shaped serial position curve in terms of willingness to respond as is shown in Table 6. Some S^s did

39 display a greater willingness to say "yes" on serial position 4 than on serial position 5. Table 6. The response bias values (Beta2) over serial position for each S Ss 1 2 3 4 5 6 Betag 1 6.32 4.77 4.34 4.80 7.48 5.82 5.59 2 6.05 4.52 3.86 4.73 7.16 5.42 5.29 3 6.09 5.24 4.30 4.84. 7.95 6.53 5.83 4 6.52 5.55 4.76 4.95 8.46 6.88 6.19 5 6.34 5.80 4.45 5.53 8.65 6.71 6.25

40 DISCUSSION A portion of the Broadbent and Gregory (1963) findings were replicated The signal detection analyses revealed that S_s showed greater sensitivity when they were not shadowing than when they were. However, the response bias index indicated that s also changed their response biases when they changed shadowing tasks. The previous findings of Broadbent and Gregory would not lead us to expect this outcome. They found no changes in response bias when Ss shifted from a shadowing to a no shadowing condition. Several procedural differences between the present experiment and the Broadbent and Gregory (1963) study could have produced the failure to replicate their findings entirely. First, it is possible that their detection task was less difficult than the recognition memory task employed here. They used as their attention dividing task the detection of a tone against a white noise background while shadowing and there might have been little reason to change the criterion for responding between the shadowing and no shadowing conditions. Also, Broadbent and Gregory did not report how accurately the S^s shadowed. In the present experiment only those trials where S^s shadowed perfectly were included in the analyses. Lastly, Broadbent and Gregory required their S^s to write down the six digits they had